Bugs for Sorcery are reported here

[bugzilla-daemon AT bugs.sourcemage.org]

http://bugs.sourcemage.org/show_bug.cgi?id=8295





------- Additional Comments From seth AT tautology.org  2005-02-22 10:39 -------
10:15  * afrayedknot finds it amusing how swoolley-work and sandalle are
suggesting things Ive already started researching (bug 8295)
10:15  * swoolley-work wonders if the channel is logged
10:15 < swoolley-work> afrayedknot: heh, I tend to do that
10:16 < swoolley-work> it's only because of the comparison to gentoo that I'm
doing -- mind you I've never downloaded it, I'm just going off what they say
themselves.
10:20 < swoolley-work> afrayedknot: what do you think of the race condition
problem that overwriting causes?  Not something to worry about?
10:20 <@afrayedknot> necessary evil?
10:21 < swoolley-work> well, it's not necessary, we can continue to dispel 
first
then install
10:21 <@afrayedknot> you cant really avoid it no matter what you do
10:21 <@afrayedknot> yea, then you have a period of time without the spell
installed at all
10:21 < swoolley-work> yeah, I was thinking that was better than having a 
period
of time when it is inconsistent.
10:21 <@dufflebunk> We could not dispel until after the new one is installed
10:21 <@afrayedknot> also, you can use a single tar command to copy the files
which minimizes the consequences of stomping on important things
10:22 < swoolley-work> not stomping important things
10:22 <@afrayedknot> as opposed to just out-right removing them
10:22 < swoolley-work> I'm thinking if somebody moved the input validation to 
a
different file in something, say in php, that for a moment if the install 
writes
over the place where it used to be, there would be no imput validation until 
it
gets to the next file.
10:23 <@afrayedknot> as opposed to not existing at all?
10:23 < swoolley-work> well, not existing at all would be better, in that case
10:23 < swoolley-work> I suppose a cost benefit would show that one behavior 
is
better though
10:24 <@afrayedknot> im not following you on how dispel + copy is better than
copy + remove left-overs
10:24 < swoolley-work> half way through install, half the spell would be old
code, half would be new code
10:24 < swoolley-work> could cause a race condition
10:25 <@afrayedknot> well, keep in mind that in either case the in-memory
programs stay the same no matter what you do
10:26 <@afrayedknot> so if you dispel them, theres a period of time where none
of the stuff it wants is around
10:26 < swoolley-work> the not-in-memory programs are where this becomes an
issue in either case
10:26 < swoolley-work> I suppose in-memory programs also might need some 
files.
10:27 <@afrayedknot> well, we currently dont address the in-memory problem at
all (try updating postfix)
10:27 < swoolley-work> maybe we can have a spell choose their flavor of update
10:27 < swoolley-work> postfix could request psuedo-atomic updates
10:28 < swoolley-work> something else could request deinstall-install updates.
10:28 < swoolley-work> default to atomic
10:28 <@afrayedknot> well, neither way would fix postfix
10:28 <@afrayedknot> you have to restart it
10:28 <@afrayedknot> or not update at all
10:28 < swoolley-work> that was a policy decision, kinda
10:29 < swoolley-work> maybe we need an exception to the policy of not
restarting services on updates for spells that it breaks.
10:29 <@afrayedknot> or putting a big fat warning to the user that they need 
to
turn it off or else it'll corrupt their mail
10:29 <@afrayedknot> or just refusing to cast until they turn it off
10:29 < swoolley-work> the latter sounds best
10:30 <@dufflebunk> You could use overlay file systems... each spell installs 
to
it's own special directory, and is overlayed onto the whole fs...
10:30 < swoolley-work> dufflebunk: I mentioned that
10:30 <@dufflebunk> but I suspect that's a lot of work... and there's no good,
reliable overlay file system around.
10:30 < swoolley-work> I think it would be a big performance hit, I imagine,
unless there's some good caching routines
10:31 <@afrayedknot> and besides, no one agrees on which filesystem to use
10:32 < swoolley-work> So, I think best situation is go ahead with 
afrayedknot's
resurrect work but have an option in DETAILS to revert back to 
deinstall/install
method?
10:32 < swoolley-work> for postfix have it fail if it is running
10:33 <@afrayedknot> i suppose, but now i have to go and re-implement the
dispel/install procedure in resurrect
10:33 < swoolley-work> then maybe later have an option to pre_install (api3)
10:33 < swoolley-work> hmm...
10:34 < swoolley-work> is it that big of a change?  It would make me feel
better, but then again, all of gentoo does it.
10:35 <@afrayedknot> its not that huge of a change
10:35 <@afrayedknot> i always considered their approach better than ours
10:35 <@afrayedknot> in that regard
10:36 < swoolley-work> in some regard :)

-- 
Configure bugmail: http://bugs.sourcemage.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.